v0.7.0

Added

• Agents Manager overlay — browse, view, edit, create, and delete agents from a TUI (Ctrl+Shift+A or /agents). List with search/filter, detail screen, field-by-field editing, model/skill/thinking pickers, templates, launch with task input.
• Chain files — .chain.md files define reusable multi-step chains with per-step frontmatter. Round-trip serializer, chain detail/edit screens, save from clarify TUI.
• Save overrides from clarify TUI — press S to persist overrides to agent frontmatter, W to save as chain file.
• Multi-select and parallel from overlay — Tab to select agents, Ctrl+R for chain, Ctrl+P for parallel builder.
• Slash commands with per-step tasks — /run, /chain, /parallel with live progress, tab-completion, and per-step task syntax: /chain scout "scan code" -> planner "analyze auth". Quotes and -- delimiter both supported. Inline config overrides ([key=val]).
• Run history — per-agent JSONL recording with recent runs on detail screen.
• Thinking level — thinking frontmatter field (off through xhigh) with picker and runtime suffix.

Fixed

• Parallel live progress now shows all concurrent tasks in real time.
• Slash commands no longer freeze input or lack progress rendering.
• /run model override (/run scout[model=gpt-4o] task) now works.
• Quoted tasks containing -- no longer split incorrectly.
• Chain first-step validation catches missing tasks in per-step mode.
• Thinking level and step-level model overrides now apply in async mode.
• mcpDirectTools env var set in async mode.
• {task} double-corruption in saved chain launches.
• Agent serializer skill normalization.
• Clarify toggle determinism across all launch paths.